The 2010s has brought a rise in the number of studies and papers discussing the role of artificial intelligence (AI) and machine learning (ML) in medicine and healthcare (AI/ML). The number of life science papers describing AI/ML rose from 596 in 2010 to 12,422 in 2019. While we are at the beginning of the AI/ML era, the expectations are high and experts foresee that AI/ML shows potential for diagnosing, managing and treating a wide variety of medical conditions1.
Indeed, AI/ML-based technologies have been shown to support several medical specialties from radiology2 and oncology3 to ophthalmology4 and general medical decision-making5. ML models have been shown to reduce waiting times6; improve medication adherence7; customize insulin dosages8; or help interpret magnetic resonance images9, among others. AI/ML- based technologies are coming to electronic health record systems and AI based notifications and messaging systems.
We classify technology like AI/ML-based if official FDA announcements, communications by the company or other publicly available information resources used the expressions ‘deep learning,’ ‘machine learning,’ ‘deep neural networks,’ ‘artificial intelligence,’ and/or ‘AI’ to describe the technology11. For simplicity, we use the term “AI/ML-based” to denote these technologies in this paper. With the increasing expertise and attention on AI/ML in the medical field, the opportunities and possible implications of its use are the topics of an ongoing debate12. A crucial element in this implementation debate is regulating such technologies.Because of the high-risk nature of these medical devices and the unknown consequences of using AI/ML for medical decision-making and data analysis, the FDA has stringent regulatory requirements for medical device licensing. Developers of AI/ML-based medical devices and algorithms have to go through rigorous processes that are time and resource consuming. This can be considered pivotal as a barrier for the introduction of AI/ML in medicine.
Before medical hardware or software is legally made available in the US market, the parent company has to submit it to the FDA for evaluation. For medically oriented AI/ML-based algorithms, the regulatory body has three levels of clearance, namely, 510(k)14, premarket approval15, and the de novo pathway16, each of which needs specific criteria to be fulfilled in order to be granted (Table 1). This process is similar to drug approvals.
Table 1 Descriptions of the types of FDA approvals for AI/ML-based medical technologies.
From: The state of artificial intelligence-based FDA-approved medical devices and algorithms: an online database
Level of FDA clearance Description
510(k) clearance A 510(k) clearance for an algorithm is granted when it has been shown to be at least as safe and effective as another similar, legally marketed algorithm. The submitter seeking this clearance must provide substantial proof of equivalence in their application. Without approval of being substantially equivalent to the other algorithm, the one pending approval cannot be legally marketed.
Premarket approval Premarket approval is issued to algorithms for Class III medical devices. The latter are those that can have a large impact on human health and as such, their evaluation undergoes more thorough scientific and regulatory processes to determine their safety and effectiveness. In order to approve an application, the FDA determines that the device’s safety and effectiveness is supported by satisfactory scientific evidence. Upon approval, the applicant can proceed with marketing the product. de novo pathway Regarding the de novo classification, it is used to classify those novel medical devices for which there are no legally marketed counterparts, but which offer adequate safety and effectiveness with general controls. The FDA performs a risk-based assessment of the device in question before approval and allowing the device to be marketed.
Table 1 Descriptions of the types of FDA approvals for AI/ML-based medical technologies.
From: The state of artificial intelligence-based FDA-approved medical devices and algorithms: an online database
TABLE Descriptions of the types of FDA approvals for AI/ML-based medical technologies. An infographic about the 29 FDA-approved, AI/ML-based medical technologies. The two main medical specialties with AI/ML-based medical innovations are Radiology and Cardiology, with 21 (72.4%) and 4 (13.8%) FDA approved medical devices and algorithms respectively. The remaining medical devices and algorithms can be grouped as focusing on internal medicine/endocrinology, neurology, ophthalmology, emergency medicine, and oncology.
The medical field of radiology is the trendsetter regarding FDA-approved medical devices and algorithms, with the introduction of AI/ML-based solutions for worldwide applied image reading software. Examples are the three algorithms for Arterys Inc., Arterys Cardio DL, Arterys Oncology DL and Arterys MICA, which are connected to the workflow Picture Archiving and Communication Systems from main vendors as Siemens Healthineers AG (Germany) and GE Healthcare (USA)20. Six out of these 21 algorithms can be applied in the field of oncology, with three focusing on mammography analyses (ProFound™ AI Software V2.1, cmTriage and TransparaTM) and three others on CT-based lesion detection (Arterys Oncology DL, Arterys MICA and QuantX). This is followed by two algorithms focusing on brain image analyses, with innovations for stroke and hemorrhage detection (ContaCT, Accipiolx, and icobrain), and six algorithms to improve image processing, with noise and radiation dosage reduction (SubtlePET, Deep Learning Image Reconstruction, Advanced Intelligent Clear-IQ Engine, SubtleMR, AI-Rad Companion (Pulmonary) and AI-Rad Companion (Cardiovascular)). Another four algorithms focusing on acute care, with two algorithms for the assessment of pneumothorax (HealthPNX and Critical Care Suite), one focusing on wrist fracture diagnosis (OsteoDetect) and the Aidoc Medical BriefCase system for triage of head, spine, and chest injuries. The final two algorithms in this specialty can be applied for cardiovascular assessments, focusing on the assessment of the heart ejection fraction (EchoMD AEF Software and EchoGo Core).
During the past decade, the use of AI/ML-based medical innovations has become ubiquitous driven by specialty needs in the clinical process from diagnosis to remote monitoring and treatment. Despite FDA approvals, end-users such as physicians, nurse practitioners must have the final word. Algorithms are continuously updated, and the FDA uses a "locked" or "adaptive" classification in order to monitor algorithms that 'learn' and change with time and use.
The authors clearly state the limitations for the FDA to accomplish this process, explaining that other organizations assess the accuracy of AI/ML-based devices.
eMurmer ID, CSD Labs GmbH), Apple Inc, being the ECG App and Apple Irregular Rhythm Notification Feature, Excel Medical Electronics, Spry Health, and Current Health, Stratoscientific, Inc. introduced the Steth IO device to analyze heart and lung sounds. BrainScope Company Inc. has introduced AI/ML for the evaluation of brain injuries. MindMotion GO (MindMaze SA), (Cantab Mobile, Cambridge Cognition Ltd), and seizure monitoring (Embrace, Empatica Srl.).
QbCheck (QbTech AB) and ReSET-O (Pear Therapeutics Inc.). With QbCheck, healthcare workers can substantiate their diagnosis or rule out attention deficit hyperactivity disorder (ADHD), enhancing objective medical decisions in psychiatry29, whereas ReSET-O can be applied for patients with Opioid Use Disorder, providing cognitive behavioral therapy as a mobile medical application for prescription use only. As a next step, the ReSET-O algorithm will be used in a randomized controlled trial, which is scheduled to start this year (April 2020)30.
The FDA has a clinical trial protocol for drugs, which may also be applied to devices and AI/ML-based devices and software.
A challenge has developed for the FDAs search algorithm, which itself will require an AI/ML-based search-based method beyond its current version.
Users of AI/ML-based software need (CME) Continuing Medical Education in each specialty.
The ACMGE as the accrediting body for medical education should supervise this as a requirement for the training of physicians
Reference source: